|
Everyone’s talking about AI, but there’s one group I particularly love hearing from: people who’ve been using and thinking about AI long before ChatGPT came along.
Avi Perez has been doing just that for 16 years. He’s the co-founder and CTO of Pyramid Analytics, a business intelligence firm well-versed in deploying AI to customers. He recently joined me for a bonus episode of the Business of Tech, where he walked me through his POV on key concepts like vectorization.
Don’t know what that is? According to Perez, you probably should; he thinks it’s the next big AI trend. From the evolution of AI to predictions on its future, here’s a rundown of our conversation.
The last 15 years of AI for BI
While many of us feel like AI just exploded on the scene, it’s nothing new to Perez. What’s he noticed from the helm of Pyramid?
Firstly, he actually can’t think of many industries outside of BI that have had AI as a central component for so long (so, don’t feel too bad about your arrival on the scene).
Secondly, until recently, his definition of AI wasn’t too fancy. He views its earlier iterations as ‘any kind of software or heuristic that can help a user to use software.’ He also differentiates AI from data science and machine learning (DSML), as AI is more about functionality – enabling people to better use a product.
“In BI, AI has been around for years. It’s been anywhere from cool, slick wizardry that helps you use very complicated software all the way through to heuristics that can work on how to build a clever formula. And in more cool and modern years, automated insights. We have now arrived at the current moment in time, which is the use of large language models to sort of superpower natural language querying,” he said.
For the record, natural language querying has been around in BI for a few years – longer than the LLMs everyone uses now – but now that we’re in a new chapter of AI, it feels a bit different.
The role of the service provider
From the service provider perspective, many of us are sensing a distinct gap between product developers who are making LLM-powered, supercharged products and our end customers. How does Perez think about the role service providers will play in closing this gap, ensuring clients implement these things responsibly and effectively?
In short, he has no doubt we’ll be important here. But when it comes to the gap between LLMs and customer implementation, he actually thinks it’s less of an LLM issue, and more of a software vendor issue – he believes it’s on the latter to close the divide.
That’s where we come in:
“The entire success of Pyramid’s gap-closing strategy is the implementation of the data strategy, which brings us all the way back to the service provider,” he said.
Once service providers support data strategies, the responsibility then goes back to the software vendor to make good use of them.
In other words…
“None of it will really go anywhere without somebody feeding Pyramid and ultimately the LLMs with good quality structure, good quality data, and a deployment that is meaningful and reasonable. And from that perspective, it all sits on top of service providers to make that happen. Otherwise, it’s garbage in, garbage out,” he said.
Perez also emphasized that it can’t just be a few random people in a company tackling the data. I think we can all agree it’ll take someone with the time and expertise to make it happen.
The opportunity for the service provider
But what’s required to pull it together? I mainly wanted to know: if Perez were building an MSP today, what would it look like?
Right now, he thinks the real lift is – you guessed it – well-structured data. With that in place, a software vendor can come along, attach their own semantic capabilities to that data structure, and then expose them to the LLM, resulting in a seamless end-user experience.
In the near future, however, he thinks upgraded LLM integrations will require the vectorization of data. In his words:
“Vectorization, in the broader sense, is the grand gluing together of public, open source information that the LLM has been trained on, that it knows about, and then bringing in something specific about the business that is not public.”
Simply put, we will need to combine an organization’s closed source, private information with an LLM that’s been trained on publicly available information.
“This is the next era, and that’s where the service providers are going to find a huge cornucopia of activity — because to make all those bits and pieces work is going to be quite a mission, and it’s not something that software solves on its own. It’s going to require people to sit down and make it work,” he said.
Prepping for the vectorization future
But we aren’t quite there yet, and Perez can’t say when we will. When we do reach this stage, however, Perez is confident vectorization will benefit from the guidance of someone who really understands the business problem andtechnical side.
Because vectorization requires so much data (more than AI already needs):
“It could be impossible to load that up into a vector database and get the performance and the scale you want at the right cost. It’s not going to be cheap, and there’s going to be something to it. So someone has to come along and carry back intelligently, understand the use case, and give not only just a prescription, but also enact that prescription to make it work,” he said.
In many ways, that’s no different from adapting to the emerging tech we’ve seen before.
What’s your take on vectorization? Up for the challenge? Feeling skeptical? As always, my inbox is open for stories, insights, questions, and more.
|
